Who Will Replace Your Best Operator?

Utilities can refill a vacancy. They cannot rehire the judgment that kept a system stable at 2 a.m. The work now is to preserve that judgment without taking the operator out of command.

By Mike Karl

*The alarm did not wake me up. A homeowner did.*

In 2003, less than six months into my first operating job at a small water utility, a homeowner called the after-hours emergency number to report a roaring sound at a tank site. When I arrived, the tank was overflowing and water was running into the storm drain.

The level transmitter had drifted out of calibration, so the booster pumps never shut off. The overflow switch activated. SCADA displayed a high-level alarm. But that alarm was not configured to shut off the booster pumps or for remote notification, so it appeared on a screen no one was watching.

I was a systems analyst learning to become an operator and recently pasted my operator-in-training license. I understood the risk but did not feel qualified for the decision in front of me. How long could the system operate with the boosters off, and what tank level protected pressure and fire-flow storage?

I found construction documents, but nothing explaining the operating levels. Our operations manager was also new. The two newest people at the utility were making an overnight decision the system could not help us make. The people who would have known had left.

We stopped the overflow and set a temporary cutoff. Finding the real answer took more than six weeks, as engineers reconstructed the logic: dead storage, operating storage, system pressure, demand, and fire-flow reserve.

All of that work produced one setpoint.

A setpoint is compressed judgment. The infrastructure had stayed. The reasoning had retired.

*The alarm did not arrive without data. It arrived without meaning.*

What Actually Leaves When An Operator Retires?

The next year, customers began calling about stained laundry and dirty water. I knew flushing was part of the answer, but not which hydrants to open or in what sequence. A senior operator understood the seasonal source, settled iron and manganese, and local flow paths. The right hydrant could contain the event to hours; the wrong one could spread it for days.

He shared the answer in a stand-up meeting. We fixed the problem. Then we went back to work. The reasoning was never captured.

Years later, an influent gate failed at a wastewater plant. The operator on shift could not stop a bypass from reaching the ocean. A more experienced operator arrived, asked about the backup hydraulic controls, and closed the gate with them.

In that case, the knowledge was documented in an SOP. It simply did not appear when the operators needed it.

These are not three technology failures. They are three forms of knowledge loss: judgment that was never written down, guidance that was written down but buried, and reasoning that could be rebuilt only through expensive engineering.

Why Have Our Existing Systems Not Solved This?

Utilities have invested heavily in SCADA, historians, GIS, asset management systems, and digital twins. Those investments matter. They help us see what is happening, understand what happened, and test what may happen next.

But they do not inherently explain why a veteran investigates one alarm and dismisses twenty others, distrusts a healthy-looking pump every spring, or chooses one hydrant over another.

SCADA holds the variables. The operator supplies the meaning.

The urgency is growing. The U.S. EPA estimates that roughly one-third of water utility operators will be eligible to retire within the next decade.¹

Across industries, APQC reports that only 8% of organizations consistently capture knowledge from departing retirees.²

A vacancy can be refilled. Decades of local pattern recognition cannot be rehired.

A utility president once told me it took about five years for a new operator to meet his organization’s experience requirements. He asked whether we could reduce that to three months. It sounded impossible until I realized he was pointing at the wrong measurement.

We cannot manufacture five years of experience in ninety days. We can stop forcing every new operator to rediscover what the last generation already learned.

*The goal is not five years of experience in ninety days. It is to stop making every new operator start from zero.*

Capture Knowledge Where It Appears

Most knowledge-transfer programs begin with the wrong request: “Write down everything you know.” That is nearly impossible. Experienced operators often do not recognize how much of their expertise has become instinctive.

The knowledge reveals itself during the work: an alarm response, a shift handoff, a near miss, a difficult startup, a troubleshooting conversation, or the moment a new operator asks a question and a veteran answers without thinking.

That is where capture should happen. The lesson should be connected to the asset, alarm, operating condition, and decision it informs. It should record where it came from, who verified it, when it was last reviewed, and what evidence supports it.

The goal is not to turn operators into technical writers. It is to make preserving judgment a small part of the work already happening.

What Should AI Be Allowed To Do?

AI can make knowledge easier to capture, organize, and retrieve. But utilities should not confuse adding AI with improving judgment.

A 2024 meta-analysis covering 106 experiments found that human-AI combinations generally performed better than people working alone, but worse than whichever participant, the human or the AI, performed better independently. The losses were especially evident in decision-making tasks.³

The study was not about water operations, but its warning belongs in the control room: adding an AI recommendation does not automatically make an operating decision safer or better. It may simply give the operator one more input to evaluate.

That distinction matters. An answer can sound convincing and still be based on an outdated procedure, the wrong operating condition, or incomplete asset history. In water operations, false confidence can be more dangerous than admitting that the answer is uncertain.

The opportunity is not to create an artificial operator. It is to give the real operator better access to the utility’s collective experience.

AI can help connect an alarm to the current SOP, retrieve similar past events, surface the reasoning behind a setpoint, capture a veteran’s explanation, and show the evidence behind a recommendation. It can help a newer operator ask better questions and help an experienced operator find relevant context faster.

But it should also make its limits visible. Operators should be able to see whether guidance came from a verified procedure, another operator’s experience, historical data, or a machine-generated inference.

In a control room, accountability cannot be ambiguous.

AI can capture, organize, surface, teach, and recommend.

The operator investigates, judges, and decides.

The operator stays in command.

Start Smaller Than A Transformation

This work does not require a massive digital program. Start with one recurring alarm or operating condition that regularly sends newer staff looking for a veteran.

Ask a veteran what they notice first, what would change their mind, and what they would never do. Connect the explanation to the SOP, trend, asset, and control narrative. Have another operator verify it, then test whether it helps a newer operator ask better questions.

That small exercise will expose the real work: outdated procedures, missing context, naming inconsistencies, untrusted data, and assumptions that exist only in someone’s head. It will also produce something useful before the organization has purchased a platform or launched a transformation.

Who Should Shape What Comes Next?

At this point, I should be transparent: I am not a neutral observer. I co-founded OCore because I believe this problem is solvable. In our work, I have watched a veteran’s voice note become searchable context tied to an asset, and a recurring alarm arrive with the procedure, equipment history, and operator explanation that gives the data meaning. That does not reproduce judgment; it gives the next operator a stronger place to begin.

But seeing the technology work has reinforced a more important lesson: no software company should define operational AI for water alone. Utilities carry public health, compliance, and safety responsibility. Operators know where context helps, where a recommendation could mislead, and where automation must stop.

*Operational AI should not be done to utilities. It should be built with them.*

That is why OCore is convening an advisory council of operators and utility leaders to shape the solution alongside us: what should be captured, how it should be verified, how sources and uncertainty should be shown, who can see it, and what must remain a human decision.

The purpose is larger than product feedback. Done well, this work can raise the standard for every platform. The opportunity is not for one company to own the industry’s judgment. It is for each utility to preserve its own while helping lift the entire industry.

None of this should feel like an exit interview. Veteran operators should feel pride in the size of what they carry, and leadership should recognize that judgment as an operational asset. Newer operators should inherit more than binders and oral tradition. They should inherit the reasoning that makes those resources useful.

Utilities should claim a seat in shaping these tools while their veterans are still on shift. The choices made now will determine whether AI becomes another layer of noise or a practical way to carry hard-earned judgment forward.

The next generation should not inherit an alarm without the meaning.

1. https://www.epa.gov/system/files/documents/2024-09/interagency-water-workforce-working-group-report-to-congress_august-2024-508-compliant.pdf
2. https://www.apqc.org/resource-library/resource/navigating-great-retirement-km-ai/html
3. https://www.nature.com/articles/s41562-024-02024-1

Mike Karl is co-founder and CEO of OCore, which is building a living operations and maintenance layer for water and wastewater utilities. He has more than 25 years of experience leading digital transformation at AECOM, Brown and Caldwell, and CH2M HILL. Karl is a licensed Water Distribution Manager, completed artificial intelligence training through MIT, and has specialized cybersecurity training from the U.S. Department of Homeland Security. He has served as a lead author of SWAN Forum industry guidance on digital twins and smart water. He is convening an advisory council of operators and utility leaders to help shape responsible operational AI. Connect with him on LinkedIn.